gh-104341: Minor Fixes in the _thread Module

[ericsnowcurrently]

This is primarily cleanup in preparation for the fix for gh-104341. However, in the process I found and fixed the following:

• PyThreadState_Delete() wasn't called at the bottom of thread_PyThread_start_new_thread()`
• the thread count (PyInterpreterState.threads.count) wasn't getting reset after fork
• the PyInterpreterState.threads.count field was effectively leaking the state of the threading module out into the broader runtime
• in thread_run(), the bootstate variable wasn't freed if its a daemon thread that exits early (due to interp finalizing)

other observations:

• unlike with bootstate, there isn't much we can do about leaking the func/args/kwargs objects in the similar case
• in threading._after_fork(), we update thread._ident but don't update thread._native_id